I guess stepwise regression would be the answer for this. However, I wonder if I can check whether it is possible to check the interaction effect of 2 independent variables.
My RQ is:
Is individual's self construal a factor for presence of meaning in a person's life to influence the life satisfaction?
I measured self construal level with a scale, which has two levels: Independent and Interdependent
I measured meaning in life with a scale, which has two levels as presence and search.
I measured life satisfaction with a scale which has a continous score at the end.
in case of 1 DV and more than one IV, multiple regression is solution. But decision is based on nature of DV. If DV is continuous variable then Multiple regression is OK, in case DV is discrete in nature then either Discriminant or Logistic regression will help. in your case, if DV is Continuous and IVs are having levels, then Multiple regression with dummy variable can be used
If I understand correctly, Life Satisfaction is the DV. Can you be more clear about the nature of this variable? I don't quite understand what "a scale which has a continous score at the end" means.
I measured the Life Satisfaction score with a scale consisting of 5 questions. Questions were 7 point likert scale from strongly disagree to strongly agree. Participants were expected to answer for all the questions and each participants' score was measured as a sum of all questions.
The general model you want is DV = IV1 + IV2 + IV1xIV2 + e, where e is just the error or residuals.
If your DV is conditionally approximately normal *, and meets the other assumptions of general linear models or ANOVA, you can construct this model as a classic 2-way anova or general linear model.
If this fails and there aren't too many different responses in the DV, ordinal regression is a good approach.
A generalized linear model may be appropriate. I'm thinking of beta regression where the DV is converted to proportion of the maximum response. Also, a gamma regression might be a good candidate.
If nothing else is appropriate, you can use a nonparametric approach. Aligned ranks with probably be the most suitable.
_______
* Technically, your DV can't be normal because it is discrete and bound (5 < DV < 35). But if the distribution approximates part of a normal distribution, the statistics should work out fine. To check this, construct the model and plot the residuals with a q-q plot or histogram. Be sure to assess the other assumptions of a general linear model.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression